Simon Michael

But I do see a significant slowdown at that commit in GHCI: 7432769d3: ``` ghci> withArgs (words "-fexamples/1000x1000x10.journal stats") main Main file : /Users/simon/src/hledger/examples/1000x1000x10.journal Included files : Transactions span :...

(gtime shows that page faults and max residency are a bit higher with the -slow builds. Perhaps increased memory pressure or differences in OS memory management accentuate the impact of...

At least, I can reliably see the speedup in GHCI when the `journalMarkRedundantCosts` line is commented in journalFinalise: ``` ghci> :main -fexamples/1000x1000x10.journal stats ... Run time (throughput) : 0.91s (1094...

Yet another benchmarking method. Apologies, but this build up of benchmarking fu is useful... ``` ~/src/hledger$ for i in `seq 1 10`; do hledger-preslow stats -f examples/1000x1000x10.journal |grep throughput; done...

I'd be open to adding this mode to the CLI. hledger-ui and hledger-web are another way to avoid re-parsing for each report.

Ledger did this back in the day, but dropped it again since it wasn't worth the added complications. I agree it would be interesting to at least try.

I'd like to get this one in today's release, and have time to take a look if needed.

I left some comments on the original PR #2087. > the workaround for now is when you use capture groups, make those regexps specific enough that at most one of...

No worries @jmtd; I went ahead with the release and noted a workaround on the issue.

Sorry you were bitten by this. Resolving the semantics and implementation are both rather tricky here. We might need to think about removing this feature until things can be clarified.